What is transcription QA, and how do you proofread a transcript?

Transcription QA is the step after the first draft, in which someone checks the transcript against the recording and the style rules before anyone relies on it. To proofread a transcript properly, you listen to the audio while you read, rather than reading the text alone. Errors cluster in predictable places, such as speaker labels, names, numbers and jargon, so the check should focus there. For higher-risk work, a second person should do a full pass.

QA checks the transcript against the audio, not only the spelling

A spell check cannot tell whether the text matches the sound. Way With Words, a transcription company, defines verification as "checking the transcript against the audio and the specification." ¹ The specification is your style guide: the choice of full, clean or edited, the tags, the timestamps and the speaker labels. A transcript can be free of typos and still break the style rules, or drop a sentence the speaker actually said.

Proofreading takes a second listen, in this order

Work through the transcript in a fixed order, so you do not miss a step:

1. Read the style rules and the speaker list before you open the audio.
2. Play the recording and read along. Stop at each speaker change, each number and each tag.
3. Check every name and technical term against your notes, and correct the spelling everywhere it appears.
4. Check the timestamps and tags. Each [inaudible] or [crosstalk] tag should sit where the words were lost.
5. Run one consistency pass for speaker labels, number style and spelling variety.
6. Run a final completeness check. The transcript should cover the whole recording, with no section skipped.

Reading the transcript alone, without the audio, misses the errors that matter most: a word the transcriber heard wrong, and a line given to the wrong speaker. Both look correct on the page.

Check the places where errors cluster first

Some errors are more likely than others, and a good review spends its time there. Way With Words says verification "should focus on predictable failure points such as speaker attribution, names, numbers, jargon, and sensitive content handling." ² Speaker attribution is first on that list for a reason. A wrong speaker label changes who said what, and a reader cannot see the mistake without the audio. Numbers and names come next, because a single digit or a misspelled surname can change the meaning and is easy to miss on a read-through.

Your own recordings may fail in other places, such as dates, drug names or case numbers. Add those to your checklist after the first few jobs, when you know which errors your transcriber makes.

Use a second pass for high-risk work

A single careful pass is enough for many meeting notes. For high-risk work, the same vendor says a second person should check the whole transcript: "Higher risk work typically requires a full second pass by another person." ³ Lower-risk work can use a structured review, but it still needs a system, not a casual read. The risk level is your call. Legal records, medical content and published quotations are usually high risk, and internal notes often are not.

How accurate is accurate enough?

Accuracy targets vary, and they measure different things. The practical exam for the Certified Electronic Transcriber, run by AAERT, requires that a candidate "demonstrate 95 percent accuracy" on a transcript. ⁴ AAERT lowered that score from 98 percent to 95 percent for exams taken on or after March 26, 2026 ⁵. That is a pass mark for an exam, not a promise about every transcript. A 95 percent transcript can still get a name or a number wrong, which is why the failure points above matter more than the average.

Research gives a sense of scale. Zayats and colleagues report that a quality checking pass improved the average transcription error rate by 5 to 20 percent. ⁶ A 1997 review reported that human transcribers had an error rate of 4 percent on spontaneous telephone conversation, which is a useful reference point for hard speech. ⁷ Each of these numbers depends on the reference transcript it was measured against, so compare like with like. Our glossary entry on word error rate explains the formula, and why the reference style changes the score.

It depends: AI drafts and human drafts need different checks

An AI draft needs the same QA as a human draft, with two extra checks. First, check the style rules. A 2025 study of dementia speech reports that Whisper "tends to omit filler words such as uh and um," which can break a verbatim requirement without anyone noticing. ⁸ Our guide on verbatim vs clean verbatim transcription explains the choices to check. Second, check the passages you can hear are difficult, such as crosstalk, accents, background noise and names. Our question on how accurate AI transcription is explains where the errors tend to appear.

A human draft needs the check too, because a single transcriber can miss things a second listener catches. The check is not a sign that the first transcriber did poor work. It is the normal step that makes a transcript dependable.

Who does transcription QA

In a larger operation, a quality reviewer or proofreader does this check. In a small job, the same person can transcribe and then review, but the review should happen after a break, with the audio again. Reviewing right after typing makes it easy to read what you meant to write rather than what was said, so the second listen is the part to protect. For more on the role, see our glossary entry on transcriptionist and the question on how to become a transcriptionist.

To set up the check before a job starts, agree on the style, the tags, the review depth and the accuracy target in writing. The question on what a good transcript looks like shows what the reviewed result should contain. For the wider subject, see our transcription overview.